https://orcid.org/0000-0002-6482-9710
Abstract
Dementia causes a substantial global economic burden, but effective treatment is lacking. Recently, studies have revealed that gamma-band waves of electrical brain activity, particularly 40 Hz oscillations, are closely associated with high-order cognitive functions and can activate microglia to clear amyloid-β deposition. Here, we found that compared with sham stimulation, applying 40-Hz high-frequency repetitive transcranial magnetic stimulation (rTMS) over the bilateral angular gyrus in patients with probable Alzheimer’s disease (AD; n = 37) resulted in up to 8 weeks of significantly improved cognitive function. Power spectral density analysis of the resting-state electroencephalography (EEG) demonstrated that 40-Hz rTMS modulated gamma-band oscillations in the left posterior temporoparietal region. Further testing with magnetic resonance imaging and TMS-EEG revealed the following: 40-Hz rTMS 1) prevented gray matter volume loss, 2) enhanced local functional integration within bilateral angular gyrus, as well as global functional integration in bilateral angular gyrus and the left middle frontal gyrus, 3) strengthened information flow from the left posterior temporoparietal region to the frontal areas and strengthened the dynamic connectivity between anterior and posterior brain regions. These findings demonstrate that modulating gamma-band oscillations effectively improves cognitive function in patients with probable AD by promoting local, long-range, and dynamic connectivity within the brain.
Introduction
Dementia is a growing global health problem that has created heavy social and economic burdens (Jia et al. 2020). Alzheimer’s disease (AD) is the most common cause of dementia and is characterized by significant cognitive decline. Its pathogenesis involves the extracellular accumulation of amyloid-β (Aβ) plaques and intracellular tau protein aggregates called neurofibrillary tangles (Ittner and Gotz 2011). Current medications for these targets are not always beneficial and have failed to hinder disease progression (Doody et al. 2014; Salloway et al. 2014). Increasing evidence suggests that Aβ plaques aggravate synapse loss and the depression of synaptic activity (Li et al. 2009; Forner et al. 2017). Specially, even soluble Aβ oligomers appearing in the very early stage of AD are toxic for synapses and impair synaptic plasticity (Querfurth and LaFerla 2010). The synaptic dysfunction and pathology that result from disturbed neurotransmission might lead to deficits in functional integration within the cognitive neural network of the brain and thus in cognition itself (Tamura et al. 2018).
Neuronal oscillations play an essential role in efficient neural communication during cognitive processing. Specifically, they facilitate the transient coupling of synchronously firing neurons that form functional neural networks (Nyhus and Curran 2010). Among all oscillation frequencies, gamma-band oscillations (30–80 Hz) are most closely associated with high-order cognitive function (Jadi et al. 2016). Previous studies have shown that gamma oscillations provide a unique and robust mechanism for effectively modulating perceptual feature binding and the encoding and retrieval of episodic memories (Hanslmayr et al. 2009; Nyhus and Curran 2010). Studies of synaptic plasticity based on spike timing showed that gamma oscillations accurately and rapidly established synchronous neuronal activity between presynaptic and postsynaptic neurons within intervals of 10–30 ms (Axmacher et al. 2006; Jutras and Buffalo 2010), which are enough to strengthen the connections between neurons and promote information transmission through Hebbian long-term potentiation (Dan and Poo 2004; Wespatat et al. 2004). Reduced gamma power accompanied by spontaneous gamma synchronization has been observed in both patients with AD and rodent models (Verret et al. 2012; Gillespie et al. 2016). A recent study has shown that 40-Hz gamma oscillations can activate microglia that clear Aβ deposition (Iaccarino et al. 2016). Furthermore, increasing the power of these oscillations can enhance attention, memory, and other cognitive functions (Jensen et al. 2007). Therefore, modulating 40-Hz gamma oscillations in the brain might constitute a promising therapy for improving cognitive function in patients with AD.
The earliest and most severe pathological changes in AD occur primarily in the memory circuit of Papez and in the posterior default mode network, which includes the medial temporal lobe, inferior parietal lobule, precuneus, and posterior cingulate gyrus (Jagust 2018). Correspondingly, the primary neurological complaints of patients with AD are deficits in episodic memory, attention, executive function, and visuospatial processing. Aβ is preferentially deposited in the cortical hubs, which have a high degree of structural and functional connectivity and play a pivotal role in functional integration (Elman et al. 2016; Grothe and Teipel 2016). The angular gyrus (AG) is one such a hub, having extensive connections with the medial temporal lobe, precuneus, dorsolateral prefrontal cortex, and superior parietal lobe. Its activity is associated with memory retrieval and formation, perceptual attention, decision-making, and manipulation (Wang et al. 2014; Sestieri et al. 2017). Therefore, improving the function of this core node might facilitate information integration within different cognitive domains for effective AD therapy (Fox et al. 2012, 2014).
Transcranial magnetic stimulation (TMS) is a noninvasive brain stimulation technology that generates a time-varying magnetic field capable of inducing electrical currents in the brain. The induced currents can elicit action potentials in neurons of the targeted brain region (Hallett 2000). TMS could create a “virtual lesion” with low-frequency stimulation as well as augment brain activity, that is, excitability, with high frequency stimulation by neuroplasticity mechanisms. In addition to these local effects, TMS can also produce distant effects in brain regions that are connected with the site of stimulation (Hallett et al. 2017). Recently, TMS has been widely used for therapy of brain disorders including depression, anxiety, insomnia, stroke, and even for AD (Burke et al. 2019).
Recent studies have shown that brain function can be modulated by rhythmic TMS pulses because they regulate brain oscillations by resetting the ongoing oscillatory activity (Brignani et al. 2008; Thut et al. 2011). Recently, gamma-rhythmic sensory stimuli (flicker and acoustic stimulation) have been used to improve cognitive function, thus providing an important advancement in the nonpharmacological treatment of cognitive impairment (Iaccarino et al. 2016, Martorell et al. 2019). Although gamma rhythmic TMS that targets local cortical areas can directly modulate local neuronal firing (Hallett, 2000), whether this technique can be used to treat AD remains unknown. Therefore, exploring whether entrainment of cortical brain rhythms via repetitive transcranial magnetic stimulation (rTMS)-evoked gamma oscillations can improve cognitive function will help us understand how dysfunctional neuronal circuits contribute to AD, which could spur the development of more effective nonpharmacological treatment protocols.
Functional magnetic resonance imaging (fMRI) is an effective tool for detecting changes in brain activity and connectivity, which can help us determine mechanisms of neuromodulation. Moreover, TMS combined with electroencephalography (TMS-EEG) is a valuable real-time method for probing changes in brain connectivity dynamics and the efficiency of information transmission at a high temporal resolution (Hallett et al. 2017; Kaarre et al. 2018; Burke et al. 2019). Thus, by combining fMRI with TMS-EEG, we can better understand functional connections and information transmission within the brains of patients with AD.
Here, we propose 2 hypotheses: 1) Can we rewire brain connections by regulating brain oscillations? 2) Does regulating gamma oscillations increase local functional connections in core brain regions and long-range functional connections between distal brain regions by enhancing synaptic plasticity, promoting information transmission, and improving cognitive function? Here, 40-Hz high-frequency rTMS over the bilateral AG was applied to determine whether rTMS can reverse cognitive decline. fMRI and TMS-EEG neuroimaging analyses were performed to determine the neurophysiological processes underlying the effects of the rTMS.
Materials and Methods
Participants
Thirty-seven patients with probable AD were recruited and randomly divided into active rTMS and sham stimulation groups (ratio: 2:1). Forty-one gender-, education level-, and age-matched healthy controls (HCs) were also enrolled. All participants were recruited from Xuanwu Hospital, Capital Medical University, China, for this study. Patients underwent a structured clinical interview and were diagnosed by 2 trained senior neurologists. The inclusion criteria were as follows: 1) right-handed men and women aged 50–85 years, 2) probable Alzheimer’s dementia diagnosed by the 2011 National Institute on Aging AD (NIA-AA) guidelines, 3) Clinical Dementia Rating (CDR) score of 0.5 or 1.0, 4) Geriatric Depression Scale (GDS) ≤8, and 5) Hachinski Ischemic Scale (HIS) <4. The exclusion criteria were as follows: 1) complications with other neurological diseases, structural brain abnormalities, or severe diseases of the heart, liver, lung, or kidney; 2) non-MRI compatible metallic implants, such as pacemakers, DBS treatment devices, or cardiac stents; 3) currently taking benzodiazepines or having a history of drug abuse; 4) a diagnosis of major depressive disorder or other severe mental illness. This study was approved by the local Medical Research Ethics Committee at Xuanwu Hospital, Capital Medical University, China. Written informed consents were provided by all participants or their surrogates (spouses or children). The trial was registered in the Chinese Clinical Trial Registry (registration No. ChiCTR1900025045). Patients who had already taken cholinesterase inhibitors for at least 6 months continued to take medications at the same dosage during treatment and follow-up. The HC group did not receive any intervention.
rTMS Treatment Paradigm
A wind-cooled figure-eight coil (70 mm diameter) was connected to an rTMS stimulator (Magstim, Co. Ltd, UK). The hot spot for the stimulation was located above the bilateral dorsal AG. The anatomical location of the dorsal AG was defined based on the AG atlas that we previously generated using both resting-state functional connectivity and task-related coactivation-based parcellation approaches (Wang et al. 2017). BrainSightTM frameless stereotaxic software (Rogue Research, Montreal, QC) was used to co-register patient heads with their structural MRI image, allowing the locations of the left and right dorsal AG to be identified accurately. All patients wore earplugs and sat in a comfortable chair in a reclined position during the treatment. All patients were treated every other day, 3 times a week for 4 consecutive weeks (a total of 12 treatment days). A daily session consisted of 30 rTMS trials (15 left AG followed by 15 right AG). The stimulus frequency was 40 Hz with 40% maximal output intensity, and a single stimulus circle contained a 2-s pulse train with a 58-s interval between pulses. Total stimulation time lasted 30 min and the total number of stimulus pulses was 2400. The coil used for sham stimulation was a 70-mm Double Air Film SHAM Coil provided by Magstim Company (Magstim, Co. Ltd, UK). The sham coil produced by the same loud sound that occurs during real stimulation. After dispersed processing, the magnetic field leaves a sense of stimulation on the scalp surface, but it does not penetrate through the skull and down into the cortex. Sham stimulation followed the same paradigm as active stimulation and helped us to better study the therapeutic effect of real magnetic stimulation.
Neuropsychological Assessment
The primary neuropsychological assessment was the Alzheimer’s Disease Assessment Scale, Cognitive Subscale (ADAS-Cog). Other cognitive measures included the Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), and CDR scale. The Neuropsychiatric Inventory, GDS, Activity of Daily Living (ADL) scale, and HIS were also used to evaluate the mental and physical status of patients. All patients completed the MMSE, MoCA, and ADAS-Cog assessments at 3 time points: pretreatment, immediately after treatment, and 8 weeks post-treatment. The control group completed neuropsychological assessments (MMSE and MoCA) once only. To exclude subjective bias, the groupings (active rTMS or sham) were blinded to the assessors. Behavioral differences before and after treatment were assessed using paired t-tests. Behavioral differences between the patients and HCs were evaluated using independent two-sample t-tests. The significant level was set at P < 0.05 with Bonferroni correction used for multiple comparisons.
Power Spectral Density Analysis of Resting-State EEG
To determine whether 40-Hz rTMS modulates gamma oscillations in the brain, we analyzed the power spectra density (PSD) of resting-state EEG using the EEG Studio toolkit (MEG Center, Cincinnati Children’s Hospital Medical Center, Ohio, USA) (Xiang et al. 2015). For each patient, resting-state EEG data that were collected with closed eyes and that did not contain any artifacts were kept for further analysis. EEGs were first segmented, then a Hanning window, zero-padded to 512 samples, was used to weight the segments before calculating the fast Fourier transform (Pan et al. 2009; Zaveri et al. 2010). Sample spectra of artifact-free 1000-ms segments were obtained for each scalp electrode and averaged over at least 3 min to calculate the PSD. The power for the theta (3–7 Hz), alpha (8–13 Hz), beta (14–30 Hz), and gamma (31–80 Hz) bands were calculated from the spectra of the scalp EEGs. Finally, paired t-tests were performed to identify differences in the patient PSDs at each frequency band before and after treatment. Significance was set at P < 0.05. Detailed analytical procedures are provided in the Supplementary Material Appendix.
VBM Analysis
A common characteristic found in patients with AD is global loss of brain volume. To test how rTMS treatment affected brain volume, voxel-based morphometric (VBM) analysis was performed to identify changes in patient brain structure after rTMS and to determine pretreatment differences between patients and controls. The structural MR images were preprocessed using a modified VBM8 toolkit (http://dbm.neuro.uni-jena.de/vbm8/) that calculated the gray matter volume (GMV) and corrected it for total intracranial volume. Longitudinal VBM analyses were used to compare GMV before and after rTMS/sham rTMS. For the longitudinal analysis, within-participant mid-way coregistration was first performed to account for between-session variance. The structural image was then segmented into gray matter, white matter, and cerebrospinal fluid and then transformed to Montreal Neurological Institute space using DARTEL-normalization. Next, the normalized gray matter images were smoothed with 6-mm full-width at half-maximum (FWHM) Gaussian kernels for statistical analyses (John and Karl 2000). For the cross-sectional comparison of structural differences between patients before treatment and controls, standard VBM procedures were employed to calculate the GMV for the controls.
Local Functional Integration Analysis
To determine how 40-Hz gamma oscillations stimulus modulate local functional integration, we used local functional connectivity density (lFCD) mapping. A specific voxel’s FCD was defined by counting the number of its functional connections greater than the threshold of 0.6. Numerous studies have demonstrated that this threshold value can effectively reduce false positive rates and is more sensitive and stable than other thresholds for revealing the functional modules of the brain (Tomasi et al. 2016; Zhang et al. 2016). To calculate the lFCD, the FC between a given voxel (x0) and each directly adjacent voxel (xi) was calculated. An FC greater than 0.6 was considered to indicate a functional connection with x0. Next, the FC between x0 and an xj that was not directly adjacent to x0, but which was directly adjacent to the first xi, was calculated, and an FC greater than 0.6 again indicated a functional connection with x0. This procedure continued until there were no more voxels with an FC greater than 0.6. For statistical analyses, the lFCD maps were normalized using z-transformation and smoothed with a 6-mm FWHM for each participant.
Long-Range Functional Integration Analysis
Functional connectivity strengths (FCSs) (also called degree centrality) were used to determine the effects of rTMS on long-range functional information integration. A voxel-wise FCS value was calculated as the average of the correlations between the voxel and all other voxels in the rest of the brain for each participant (Wu et al. 2016; Liu et al. 2018). To calculate the FCS, the functional connectivity of a given voxel with all the other voxels of the brain was first calculated using Pearson’s correlation coefficients. Then, a threshold of 0.25 was used to remove the weak connections arising from signal noise. This threshold has been shown to exclude false connections and keep the intrinsic connectivity patterns (Wu et al. 2016; Liu et al. 2018). An FCS map for each participant was obtained by averaging the correlation coefficients higher than the threshold across the whole brain and then transformed to Z-scores to improve normality. In terms of graph theory, this FCS metric is called the “degree centrality” of weighted networks (Buckner et al. 2009; Zuo et al. 2012). The FCS map was finally spatially smoothed with a 6-mm FWHM Gaussian kernel and subjected to statistical analyses.
Seed-Based FC Analyses
To identify whether the brain areas responding to rTMS treatment showed altered functional connectivity, seed-based whole-brain resting-state FC analyses were performed with the seed regions being the brain areas that exhibited altered GMV, lFCD, and FCS. The Pearson correlation coefficients between the mean time series of each seed region and the mean time series of each voxel in the whole brain were computed and converted to z values using Fisher’s z transformation to improve normality.
Statistical Analyses of MRI Data and Correlation Analyses
First, GMV differences before and after rTMS treatment were identified between the controls and patients using two-sample t-tests. The mean GMVs for brain areas showing significant differences were extracted and then subjected to paired t-tests to determine before/after treatment differences. The significant level was set at P < 0.05 with Bonferroni correction. We also performed before/after paired t-tests for the lFCD and FCS. The significant level for all the voxel-wise whole-brain statistical analyses was determined using a cluster-level Monte Carlo simulation (5000 times), with a corrected threshold of P < 0.05 (cluster-forming threshold at voxel-level P < 0.001). To compare the lFCD and FCS between controls and patients before or after rTMS treatment, we first extracted the mean lFCD and FCS for the brain regions in which differences were found before and after rTMS. Then, we performed two-sample t-tests on these values between controls and patients, before and after rTMS. Significance was set at P < 0.05 with Bonferroni correction.
To investigate the relationship between brain measurements and clinical characteristics, we performed correlation analyses. Significance was set at P < 0.05.
Dynamic Connectivity Analysis of TMS-EEG
To explore how the impact of rTMS on the dynamic effective FC between AG and other brain areas, high temporal resolution TMS-EEG data were analyzed within the 3–80 Hz frequency range. TMS-evoked disturbances in the EEG data were labeled and the time points corresponding to each label were marked. Then, the data corresponding to 500 ms before and 1000 ms after the markers were extracted (1500 ms in total). Eight-rate downsampled segmentation was performed to reduce the calculation load (Li et al. 2016). A time-varying adaptive multivariate autoregressive (AMVAR) model and an adaptive directed transfer function (ADTF) were used to calculate dynamic connectivity (Zhang et al. 2017). The time-varying coefficient matrices of the AMVAR were resolved using the Kalman filter algorithm, then the order of the model was determined by the Akaike information criterion. The ADTF value was used to construct the time-varying networks and examine dynamic information processing during the TMS-evoked disturbances in the EEG (Wilke et al. 2007). Detailed explanations of the procedures for the dynamic EEG analyses are included in the Supplementary Material Appendix.
Data Availability
All data reported in the results can be provided by request to the corresponding authors from a qualified investigator for noncommercial use (sharing of some data is subject to restrictions according to participant consent and data protection legislation).
Results
Demographic and Clinical Characteristics
Patients in the active rTMS, sham, and HC groups did not differ in terms of age (F = 3.04, P = 0.054), gender (χ2 = 2.52, P = 0.28), or education level (F = 1.07, P = 0.35). Furthermore, no significant differences in the age of onset (t = −1.91, P = 0.06), disease duration (t = 0.30, P = 0.77), or CDR (t = 0.78, P = 0.44) were found between the active rTMS and sham groups (see Table 1 for details).
Demographic information for HCs and patients with mild probable AD
| Mild AD | HC | ||
|---|---|---|---|
| Active rTMS | Sham rTMS | ||
| Age (year) | 67.28 ± 7.74 | 72.08 ± 7.30 | 66.43 ± 6.44 |
| Gender (female/male) | 12/13 | 4/8 | 24/17 |
| Education level (years) | 10.44 ± 3.33 | 11.66 ± 2.53 | 11.56 ± 3.33 |
| Age of onset (years) | 64.16 ± 7.40 | 69.16 ± 7.57 | — |
| Disease duration (years) | 3.12 ± 2.20 | 2.91 ± 1.08 | — |
| CDR | 0.84 ± 0.55 | 0.71 ± 0.26 | — |
| Mild AD | HC | ||
|---|---|---|---|
| Active rTMS | Sham rTMS | ||
| Age (year) | 67.28 ± 7.74 | 72.08 ± 7.30 | 66.43 ± 6.44 |
| Gender (female/male) | 12/13 | 4/8 | 24/17 |
| Education level (years) | 10.44 ± 3.33 | 11.66 ± 2.53 | 11.56 ± 3.33 |
| Age of onset (years) | 64.16 ± 7.40 | 69.16 ± 7.57 | — |
| Disease duration (years) | 3.12 ± 2.20 | 2.91 ± 1.08 | — |
| CDR | 0.84 ± 0.55 | 0.71 ± 0.26 | — |
No differences among the 3 groups were found for age (F = 3.04, P = 0.054), gender (χ2 = 2.52, P = 0.28), or education level (F = 1.07, P = 0.35). Additionally, no differences in the age of onset (t = −1.91, P = 0.06) disease duration (t = 0.30, P = 0.77), or CDR (t = 0.78, P = 0.44) were found between the active rTMS and sham rTMS groups.
Demographic information for HCs and patients with mild probable AD
| Mild AD | HC | ||
|---|---|---|---|
| Active rTMS | Sham rTMS | ||
| Age (year) | 67.28 ± 7.74 | 72.08 ± 7.30 | 66.43 ± 6.44 |
| Gender (female/male) | 12/13 | 4/8 | 24/17 |
| Education level (years) | 10.44 ± 3.33 | 11.66 ± 2.53 | 11.56 ± 3.33 |
| Age of onset (years) | 64.16 ± 7.40 | 69.16 ± 7.57 | — |
| Disease duration (years) | 3.12 ± 2.20 | 2.91 ± 1.08 | — |
| CDR | 0.84 ± 0.55 | 0.71 ± 0.26 | — |
| Mild AD | HC | ||
|---|---|---|---|
| Active rTMS | Sham rTMS | ||
| Age (year) | 67.28 ± 7.74 | 72.08 ± 7.30 | 66.43 ± 6.44 |
| Gender (female/male) | 12/13 | 4/8 | 24/17 |
| Education level (years) | 10.44 ± 3.33 | 11.66 ± 2.53 | 11.56 ± 3.33 |
| Age of onset (years) | 64.16 ± 7.40 | 69.16 ± 7.57 | — |
| Disease duration (years) | 3.12 ± 2.20 | 2.91 ± 1.08 | — |
| CDR | 0.84 ± 0.55 | 0.71 ± 0.26 | — |
No differences among the 3 groups were found for age (F = 3.04, P = 0.054), gender (χ2 = 2.52, P = 0.28), or education level (F = 1.07, P = 0.35). Additionally, no differences in the age of onset (t = −1.91, P = 0.06) disease duration (t = 0.30, P = 0.77), or CDR (t = 0.78, P = 0.44) were found between the active rTMS and sham rTMS groups.
Compared with controls, patients with probable AD exhibited poorer cognitive function before treatment, as evidenced by lower MMSE and MoCA scores (Fig. 1 and Table S1). Analysis showed significant improvements in cognitive functions (increased MMSE and MoCA scores and reduced ADAS-Cog scores) in the patients after active rTMS treatment. Improved clinical symptoms remained at the 8-week follow-up (Fig. 1 and Table S1). In contrast, we did not find any significant changes in cognitive function (i.e., MMSE, MoCA, and ADAS-Cog scores) for the sham group (Fig. 1 and Table S1). These findings indicate that 40-Hz rTMS was able to reverse declines in cognitive function and sustain the reversal for several months.
Scores on the neuropsychological scales before, immediately after, and 8 weeks after rTMS/sham treatment. *P < 0.05.
Changes in Gamma-Band PSD
Because EEG signals in 3 participants were faulty, data from only 22 patients were used for the resting-state EEG analysis for the active rTMS group. The PSD analyses revealed significant treatment effects at electrode T5 (P = 0.046) and Cz (P = 0.012), both of which showed significantly increased gamma oscillation power after active rTMS treatment (Fig. 2A,C). We found no significant differences at other frequency bands in the active rTMS group. Additionally, we found no significant changes in PSD for the sham group (all patients, n = 12; Fig. 2B).
Topography of gamma band (30–80 Hz) activity and the t-map. (A) The gamma PSD maps before and after rTMS treatment in patients with probable AD. Before/After difference maps are also shown. (B) Same as A, but for the sham stimulation group. (C) A significant increase in gamma-band PSD at electrodes T5 and Cz were found after rTMS treatment. No significant differences in whole-brain average PSD were observed.
Pattern of GMV Changes in Patients before and after rTMS Treatment Compared with Healthy Controls
Compared with the control group, patients with probable AD exhibited significantly lower GMV in the lateral and medial temporal cortex, inferior parietal lobule, precuneus, thalamus, and insula before and after rTMS treatment (Fig. 3A). These GMV patterns were similar and highly overlapped (Fig. 3B). These overlapping regions of low GMV did not differ in the active rTMS group but continued decreasing in the sham stimulation group (Fig. 3C), suggesting rTMS might help prevent brain atrophy (Fig. 3C).
Changes in GMV before and after rTMS/sham stimulation compared with GMV in controls. (A) Compared with controls, patients with AD exhibited lower GMV before and after rTMS in temporal, parietal, subcortical areas, and the hippocampus. (B) The regional patterns of low GMV in patients with probable AD were similar before and after rTMS. The ratio is defined as the number of voxels in brain areas with GMV loss divided by the total number of voxels in brain areas with GMV loss found in probable AD patients before and after rTMS treatment. (C) Patients with probable AD exhibited regions before and after rTMS that contained lower GMV than control participants and which overlapped with each other. These overlapping regions of low GMV did not differ in the amount of GMV before and after rTMS treatment, but they did differ after sham stimulation, with GMV continuing to decrease. *P < 0.05.
Changes in Local Functional Integration after rTMS Treatment
Local functional integration was characterized using lFCD. lFCD analysis showed that rTMS treatment enhanced local functional integration with the bilateral AG, which normalized the function across the 2 areas. Furthermore, we observed altered lFCD for the left AG after rTMS, and the degree of this effect was negatively correlated with the executive function scores on the ADAS-Cog. However, the sham treatment group only exhibited significantly increased lFCD in the right AG (Fig. 4). These results may indicate that rTMS mainly modulated the local functional integration of the left AG, which helped improve cognition in patients with probable AD.
The FCD before and after rTMS/sham stimulation. In the active rTMS group, FCD analysis revealed enhanced local functional integration of the bilateral AG. In the sham stimulation group, FCD was slightly increased in bilateral AG, but only the increase in the right AG was statistically significant. The change in local FCD in the left AG after treatment was negatively correlated with the executive function score on the ADAS-cog. AG_L, left AG; AG_R, right AG; *P < 0.05.
Changes in Long-Range Functional Integration after rTMS Treatment
FCS was employed to examine how rTMS affected long-range functional integration. In the rTMS group, FCS increased significantly in the bilateral AG (AG_L and AG_R) and left middle frontal gyrus (MFG_L) after treatment. In contrast, FCS did not change after sham stimulation (Fig. 5A). Moreover, the degree to which FCS increased after rTMS stimulation in the bilateral AG and left MFG significantly and negatively correlated with the change in ADAS-Cog scores. This suggests that the degree of FCS increase in these areas can predict clinical improvements in patients with probable AD (Fig. 5B).
FCS before and after rTMS/sham stimulation. (A) FCS analysis revealed that the whole-brain functional connection strength of the bilateral AG and left MFG increased after rTMS treatment. Although the same areas showed a slight increase after sham stimulation, it was not significant. (B) Correlation analysis found that enhanced FCS after rTMS treatment was negatively correlated with reduced ADAS-Cog scores, suggesting that it led to clinical cognitive improvements. AG_L, left AG; AG_R, right AG; FCS, functional connection strength; *P < 0.05.
Changes in Functional Couplings after rTMS Treatment
We performed resting-state functional connectivity analyses to determine whether rTMS can enhance functional interactions between brain areas in patients with probable AD. Compared with the control group, patients before treatment exhibited weaker FCS between the left AG (AG_L) and both the left superior frontal gyrus (SFG_L) and right inferior frontal gyrus (IFG_R) (Fig. 6). After rTMS treatment, the FCS was normalized to the level of controls. Furthermore, the FCS between the left AG and left SFG in the rTMS group correlated negatively with their scores for memory function on the ADAS-cog. As expected, no significant changes in functional connectivity between these areas were observed in the sham group (Fig. 6).
Functional connections (FCs) before and after rTMS/sham stimulation. (A) In the rTMS group, we found fewer FCs between the left AG and the left SFG and between the left AG and the right IFG in the patients than in HCs. After treatment, the number of FCs in the patients increased and the difference with controls was reduced. The same brain regions showed no change in the number of FCs after sham stimulation. (B) The degree to which FCs increased between the left AG and the left SFG was negatively correlated with memory function scores on the ADAS-cog. FC, functional connections; *P < 0.05.
Changes in Dynamic Connectivity Via TMS-EEG
We found significant differences in time-varying EEG-network dynamic connections before and after treatment in the rTMS group. Specifically, the analysis revealed increased information outflow throughout the observation period. In contrast, this increased outflow was rarely observed in the sham group. In the rTMS group, the increased information outflow originated from the left temporal–parietal region (T5). At an early stage (60–180 ms), T5 induced an increase of information outflow to frontal regions (F3, FZ, F4). At the middle stage (300–540 ms), information outflow from T5 to frontal regions continued to increase and additional increases in connections were observed going to the anterior and posterior midline brain areas. At the late stage (660–900 ms), information output was less than in the middle stage but remained primarily from T5 to bilateral frontal areas. In the sham stimulation group, only local connections between the frontal lobe and the central area increased in the late stage. Thus, rTMS was characterized by a poststimulation increase of information output from the left posterior temporoparietal region to the frontal region. In addition, increased connections between anterior and posterior midline brain regions were also prominent (Fig. 7A,B).
Topography of dynamic connections revealed by TMS-EEG (3–80 Hz) from a paired t-test map comparing rTMS and sham groups. The topographies represent the significantly enhanced dynamic connectivity after rTMS treatment (A) and sham stimulation (B). The red line indicates enhanced functional connectivity, and the black triangle indicates one-way information outflow. The yellow line without a triangle indicates bilateral information flow.
Discussion
In this study, we reported that gamma-band 40-Hz rTMS improved cognitive function in patients with probable AD for at least 8 weeks by modulating gamma oscillations in the brain. Using combined MRI and EEG techniques, we further found that rTMS 1) strengthened the gamma-band power of the left temporoparietal cortex, 2) prevented GMV loss, 3) increased local functional integration in bilateral AG, 4) enhanced long-distance functional connections between the left AG and both the left SFG and the right IFG, and 5) promoted more robust information output from the left posterior temporoparietal region to frontal areas, as well as strengthening the dynamic connectivity between anterior and posterior brain regions. Notably, the enhanced local and long-range functional integrations and the enhanced FC were closely associated with improvements in clinical symptoms including executive function and memory. These results demonstrated that modulating gamma-band oscillations can effectively improve cognitive function in patients with probable AD by increasing local, long-range, and dynamic connectivity, thereby promoting information flow and integration. Furthermore, well-being was maintained in all patients, and none reported any adverse effects during the rTMS treatment.
Because the processing capacity of sensory systems is limited, attention helps select stimulus features related to behavioral goals from a rich and noisy environment. Attended visual and somatosensory stimuli trigger stronger gamma-band responses than unattended stimuli (Bauer et al. 2006). Tallon-Baudry et al. (1998) found that active maintenance of abstract visual shapes in short-term memory was related to enhanced gamma-frequency power at occipital and bilateral temporal lobes. Furthermore, studies have shown that gamma-frequency activity during encoding can predict successful formation of long-term memory (Jensen et al. 2007; Sederberg et al. 2007). Our study found that the power of gamma oscillations in the left temporoparietal region increased in patients with mild AD after treatment with 40-Hz rTMS. This was accompanied by improved attention, executive function, and memory function. Our findings provide additional evidence that gamma oscillations are closely related to cognitive functions and new evidence that inducing these oscillations artificially can improve cognitive impairment.
The AG, particularly in the left hemisphere, serves as a memory buffer for intentional maintenance of episodic information until the execution of an action (Kalpouzos et al. 2010). Additionally, the SFG plays a functional role in working memory (Alagapan et al. 2019). Coactivation of dorsal AG and the SFG has been reported to be involved in memory retrieval (Nelson et al. 2010). This is consistent with our findings that increased functional connectivity between the left AG and left SFG correlated with improved memory function after active rTMS treatment. More concretely, the AG seems to act as a cross-modal integrative center through which multisensory information converges. Based on previous expectations, knowledge, and intended action, the AG could give sense and meaning to specific events within a contextualized condition. These findings highlight the integrative role of the AG in supporting fast communication across brain regions (Liang et al. 2013; Tomasi et al. 2013). A recent study reported that patients with AD showed significant decreases in FCS in many hub regions, suggesting disrupted hub-related connectivity and modular integrity in AD (Dai et al. 2015). The AG exhibits significant hypometabolism in patients with AD, and its dysfunction is related to cognitive impairment (Gaubert et al. 2017). Furthermore, resting-state fMRI and positron emission tomography (PET) studies have consistently identified the AG as a central parietal node of the DMN that is associated with AD (Banks et al. 2018). Our study found increased lFCD in the bilateral AG after active rTMS treatment, suggesting that enhancing local connections and integration of core nodes might play a vital role in facilitating information processing.
In addition to interactions within local regions, complex cognition often requires long-distance integration across the brain network. Studies have suggested that long-range connections provide quick links between remote brain regions within networks and play crucial roles in supporting human cognitive function (Achard et al. 2006; Bullmore and Sporns 2012). A drop in long-range connections might disrupt functional integration between different brain regions and underlie the cognitive impairments observed in AD. This theory was supported by findings that the most significant AD-related FCS decreases appeared in the 100–130 mm range and that significant correlations existed between the long-range hub FCS values and general cognitive performance (MMSE and MoCA) in patients with AD (Dai et al. 2015). EEG, PET, and MRI studies have all documented the large-scale network disconnection between anterior and posterior areas of the brain (Bullmore and Sporns 2012). A general decrease in MEG coherence values was observed in patients with AD within all frequencies, although the relative coherence loss was highest for the long-distance frontoparietal coherence measures (Berendse et al. 2000). After rTMS treatment, we found significantly increased FCS in the bilateral AG and left MFG of patients with probable AD. The increase in FCS was significantly correlated with improved cognitive function, suggesting that distant connectivity also plays an essential role in cognitive processing.
A recent investigation surprisingly reported that optogenetic driving of 40-Hz gamma oscillations in PV-positive interneurons markedly reduced toxic levels of amyloid-β and plaques in the hippocampus of a transgenic mouse model of AD (Iaccarino et al. 2016). Moreover, using 40-Hz auditory stimulation to drive gamma oscillations of neural activity in the auditory cortex and hippocampal CA1 reduced amyloid levels and improved memory in animal models of AD (Martorell et al. 2019). These data strongly suggest that neural circuits producing gamma rhythms are essential for amyloid clearance that occurs via increased microglial activity, which reduces synaptic toxicity (Iaccarino et al. 2016). Strengthening synaptic connections in a neural activity-dependent manner is vital for oscillation-related neural cognition. If a neuron receives input from several other neurons, this drive is enhanced when the spiking inputs are coincident because synchronization enables the postsynaptic potentials to integrate different inputs (Salinas and Sejnowski 2001). Given that the duration of an excitatory postsynaptic potential is approximately 10 ms (Williams and Johnston 1991), gamma-frequency oscillations lasting 10–30 ms provide a “tighter” synchronization than oscillations at lower frequency bands. This principle enables a group of neurons to exert a stronger drive when they are synchronized with the downstream areas, so as to enhance information transmission and communication between different regions of the brain (Fries 2005). Thus, increasing gamma activity in the brain is a promising therapeutic strategy for treating patients with AD.
Presenting rhythmic sensory stimuli in visual, auditory, and somatosensory modalities is established methods of modulating brain oscillations (Herrmann et al. 2016). To date, only a few studies have reported on the therapeutic efficacy of gamma sensory stimulation in small samples of patients with AD. In a pilot study, Cortes and colleagues investigated whether entraining the somatosensory system with a sound-driven vibrotactile stimulation paradigm at 40 Hz could improve cognition. The results showed that after 40-Hz sensory stimulation (rather than control stimulation), alertness, cognition, and short-term memory at all AD disease stages were improved (Clements-Cortes et al. 2016). In a longitudinal case study (>3 years), the same research group used a similar gamma stimulation paradigm for an 89-year-old female patient with AD and found that the gamma stimulation helped her maintain her cognitive abilities (Clements-Cortes et al. 2017). Compared with rhythmic sensory stimulation, transcranial alternating current stimulation (tACS) and rTMS directly stimulate relevant cortical areas to entrain endogenous brain oscillations, resulting in more potent effects on brain activity. In a randomized, double-blind, sham controlled, crossover pilot study, a single 60-min treatment with exposure to gamma-tACS over Pz (an area overlying the medial parietal cortex and the precuneus) can improve memory performances of patients with mild cognitive impairment due to Alzheimer’s disease (MCI-AD) (Benussi et al. 2021). Another pilot study showed cognitive improvement induced by brain exercises can be enhanced by repeated and simultaneous application of tACS at 40 Hz in an MCI or mild to moderate dementia population (Kehler et al. 2020). Recently, a protocol for a randomized controlled clinical trial using cumulative 40-Hz tACS to treat patients with mild AD has been published (Xing et al. 2020); however, this project is still ongoing and the results of this clinical trials with cognitive endpoints have not been obtained. In our study, we first reported using 40-Hz gamma-band rTMS to treat patients with mild AD. When performing cognitive tasks, brain regions interact through complex and dynamic neural connections (Uddin et al. 2014). In patients with AD, information synchronization from the parietal to frontal regions was weaker than what was found in older people without AD (Babiloni et al. 2009). In our study, treating AD with 40-Hz gamma-band rTMS led to increased dynamic connectivity that originated from the left posterior temporoparietal region and extended to the frontal areas, as well as increased long-distance information transmission between anterior and posterior brain regions. Our findings thus demonstrate that 40-Hz rTMS could play an essential role in treating dysfunctional synaptic transmission by promoting gamma activity synchronization and providing a novel noninvasive, nonpharmacological therapeutic approach for treating AD.
Recent findings indicate the possibility of predicting cognitive deterioration with a panel of TMS-derived measures. The enhanced brain excitability and plasticity seem to play a role in counteracting the decline of cognitive abilities in older people that results from neuronal loss and vascular injury (Giuseppe et al. 2017). In our study, although the HIS scores for all participants were below 4, the so-called normal-appearing white matter analysis may reveal “soft” or “silent” microvascular disease in both normal brain aging and neurodegenerative disorders (Wang et al. 2013; Maniega et al. 2015). The damage of microstructure integrity in normal-appearing white matter that results from ischemic interruption leads to functional changes in intracortical excitatory circuits, as well as cortical–subcortical circuits. This change can be evaluated using noninvasive brain stimulation techniques, such as TMS-related short-latency afferent inhibition or intracortical facilitation, as well as the transcranial Doppler-related index (Vinciguerra et al. 2019, 2020; Cantone et al. 2020), which might help us refine our tests in future research.
Our study has some limitations. First, the number of patients with probable AD who were enrolled in this study was not large. More patients from multiple centers need to be included in future studies. Second, several patients were medicated while receiving rTMS treatment, and thus, a potential cumulative effect of magnetic stimulation and medication-related effects cannot be excluded. Third, outcomes were only measured immediately after treatment and at an 8-week follow-up. In our next study, longer follow-up assessments, such as 6 months and 1 year after treatment, will be conducted. Finally, patients in sham groups were slightly older than those in the rTMS group, which might have contributed to the differences in outcomes between the groups.
In conclusion, this study provides preliminary evidence for the clinical efficacy of 40-Hz rTMS over the AG for treating cognitive impairment. Imposing rhythmic gamma oscillations can increase local connectivity within this hub region, enhance long-distance connections between the frontal and parietal areas, and strengthen dynamic information transmission of large-scale networks to improve cognitive function. Our findings provide a neurophysiological basis for using 40-Hz neuromodulation therapy to treat cognitive impairment and can help facilitate a better understanding of cognitive disorders and potential treatments.
Funding
National Key R&D Program (No. 2018YFC1314500, 2018YFC1314504); National Natural Science Foundation of China (No. 62176044, 81701297, 81801124, 82071483); Sichuan Science and Technology Program (2021YJ0186).
Notes
We thank our volunteers for their participation in this study, and the radiographers/technologists at the Brain Imaging Centre of Xuanwu Hospital. We would also like to thank the therapists at the neuromodulation therapy clinic of Xuanwu Hospital and the scale assessors at the neuropsychological examination room of Xuanwu Hospital. Lastly, we want to thank Adam Phillips, PhD, from Liwen Bianji, Edanz Editing China (www.liwenbianji.cn/ac), for editing the English text of a draft of this manuscript. Conflict of Interest: None of the authors has any conflict of interest to disclose.
References
Author notes
Chunyan Liu, Tao Han, and Zhexue Xu contributed equally to this work.
